Published in "Botanical Journal of the Linnean Society 192(4): 656–674, 2020" which should be cited to refer to this work.

Systematics of (): phylogenetic relationships based on nuclear gene and partial plastome sequences

TALITA MOTA MACHADO1*, , ORIANE LOISEAU2, MARGOT PARIS3, ANNA WEIGAND4, LEONARDO M. VERSIEUX5, JOÃO RENATO STEHMANN1, CHRISTIAN LEXER6 and NICOLAS SALAMIN2,

1Departamento de Botânica, Instituto de Ciências Biológicas, Programa de Pós-Graduação em Biologia Vegetal, Universidade Federal de Minas Gerais, Belo Horizonte, CEP 31270–901, MG, 2Department of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland 3Department of Biology, Unit Ecology and Evolution, University of Fribourg, Fribourg, Switzerland 4Department of Systematic and Evolutionary , University of Zurich, 8008, Zurich, Switzerland 5Laboratório de Botânica Sistemática, Departamento de Botânica e Zoologia, Centro de Biociências, Universidade Federal do Rio Grande do Norte, Natal, CEP 59078–970, RN, Brazil 6Department of Botany and Biodiversity Research, Faculty of Life Sciences, University of Vienna, Rennweg 14, 1030 Vienna, Austria

Vriesea is the second largest in , the most diverse subfamily of Bromeliaceae. Although recent studies focusing on Tillandsioideae have improved the systematics of Vriesea, no consensus has been reached regarding the circumscription of the genus. Here, we present a phylogenetic analysis of core Tillandsioideae using the nuclear gene phyC and plastid data obtained from genome skimming. We investigate evolutionary relationships at the intergeneric level in Vrieseeae and at the intrageneric level in Vriesea s.s. We sampled a comprehensive dataset, including 11 genera of Tillandsioideae and nearly 50% of all known Vriesea spp. Using a genome skimming approach, we obtained a 78 483-bp plastome alignment containing 35 complete and 55 partial protein-coding genes. Phylogenetic trees were reconstructed using maximum-likelihood based on three datasets: (1) the 78 483 bp plastome alignment; (2) the nuclear gene phyC and (3) a concatenated alignment of 18 subselected plastid genes + phyC. Additionally, a Bayesian inference was performed on the second and third datasets. These analyses revealed that Vriesea s.s. forms a well-supported clade encompassing most of the of the genus. However, our results also identified several remaining issues in the systematics of Vriesea, including a few species nested in and http://doc.rero.ch Stigmatodon. Finally, we recognize some putative groups within Vriesea s.s., which we discuss in the light of their morphological and ecological characteristics.

ADDITIONAL KEYWORDS: Atlantic Forest – chloroplast – epiphytes – genome skimming – – Neotropics – next-generation sequencing – Tillandsioideae.

INTRODUCTION family has been suggested to be the result of rapid speciation and adaptive radiations mainly triggered by Bromeliaceae are a large, nearly entirely Neotropical the evolution of key innovations that enabled species family with a wide geographical distribution to colonize new adaptive zones (Givnish et al., 2011, ranging from the southern United States to northern 2014; Silvestro, Zizka & Schulte, 2014). Morphological Patagonia in ; a single species occurs in features, such as tank-forming leaves, epiphytism, Africa (Smith & Downs, 1974). Diversification in the water- and nutrient-absorbing leaf trichomes and CAM photosynthesis, have allowed bromeliads to *Corresponding author. E-mail: [email protected] colonize highly heterogeneous environments (Benzing,

1 2000; Silvestro et al., 2014). Consequently, bromeliads Butcher & Gouda, cont. updated; Palma-Silva et al., occur from mesophytic forests to xerophytic rock 2016; Schütz et al., 2016). outcrops, from sea level (e.g. restingas in Atlantic Vriesea, the fourth largest genus in Bromeliaceae Forest) to the top of mountains in the Andes and the (Gouda et al., cont. updated), highlights well the Brazilian Shield (Gilmartin, 1973; Krömer, Kessler & problems faced by current phylogenetic studies. In Herzog, 2006; BFG, 2015, 2018). The wide variation the genus, 88% of the species in the genus occur in in morphological characters and the low rates of Brazil, and 84% of all Vriesea spp. are endemic to the molecular evolution observed in Bromeliaceae (Smith country (BFG, 2015, 2018). They are distributed from & Donoghue, 2008; Maia et al., 2012) make this family xeric to mesic environments (Versieux & Wendt, 2007; of c. 3587 species in 75 genera (Butcher & Gouda, cont. Versieux, 2008; Costa, Gomes-da-Silva & Wanderley, updated) a challenging group for taxonomists. Because 2014; Machado, Forzza & Stehmann, 2016) and are a of these characteristics, the delimitation of species, or conspicuous element of the Atlantic Forest (Martinelli sometimes even genera, has proved difficult (Palma- et al., 2008). Some species reach a western inland Silva et al., 2016). distribution, growing as epiphytic, terrestrial or Based on molecular phylogenetic analyses, eight rupicolous on inselbergs and in rocky savanna- subfamilies of Bromeliaceae, the circumscription of like habitats (Versieux & Wendt, 2007; Versieux et al., which was usually based on only a few plastid markers, 2008; Costa et al., 2014). The geographical distribution are currently recognized in the literature (Terry, Brown of species varies from wide ranges covering different & Olmstead, 1997a, b; Horres et al., 2000; Crayn, habitats to microendemics that are particularly Winter & Smith, 2004; Barfuss et al., 2005; Givnish frequent in mountainous environments (Versieux & et al., 2007). In the last decades, many phylogenetic Wendt, 2007; Costa et al., 2014; Machado, Forzza & studies have improved our understanding of the Stehmann, 2016). Due to the restricted distributions of evolutionary history of these subfamilies, but species many Vriesea spp. and the continuous loss of habitat, representation has been highly heterogeneous among the genus is the second most endangered in Brazil studies (Escobedo-Sarti et al., 2013; Palma-Silva et al., when considering the absolute number of endangered 2016). For instance, among the subfamilies with the plant species (Martinelli et al., 2013). The wide climatic highest number of published phylogenetic studies tolerance of Vriesea is coupled with morphological ( and Tillandsioideae), Tillandsioideae variation (Costa et al., 2014; Costa, Gomes-da-Silva & are also the subfamily with the most scattered and Wanderley, 2015). Phenotypic variability occurs also at least connected sampling because most publications the intraspecific level, making the delimitation of taxa analysed only small clades (Escobedo-Sarti et al., complicated, and leading to the recognition of several 2013). As a result of the low rates of substitution in species complexes (Almeida et al., 2009; Gomes-da- plastid regions, the tree topologies are often poorly Silva & Costa, 2011; Versieux, 2011; Neves et al., 2018). resolved not helping in the circumscription of some The polyphyletic condition of Vriesea and the genera in Bromeliaceae (Smith & Donoghue, 2008; difficulty in recognizing unique morphological Maia et al., 2012; Palma-Silva et al., 2016); non-natural characters to circumscribe the genus have been known genera are frequently reported in Bromeliaceae for over two decades (Terry et al., 1997a; Barfuss (Barfuss et al., 2005; Evans et al., 2015; Schütz et al., et al., 2005; Givnish et al., 2011; Gomes-da-Silva et al., 2016; Maciel et al., 2018) and only 30% of the currently 2012; Costa et al., 2015; Gomes-da-Silva & Souza- http://doc.rero.ch recognized genera are monophyletic (Escobedo-Sarti Chies, 2017). Recent studies using morphological et al., 2013). To overcome these problems, recent characters and molecular markers have elucidated studies have increased the amount of analysed data the intergeneric relations in Tillandsioideae and either by increasing the number of molecular markers helped to improve the circumscription of Vriesea and their phylogenetic informativeness (e.g. more (Barfuss et al., 2016; Gomes-da-Silva & Souza-Chies, quickly evolving nuclear regions or even AFLP) or 2017). However, conflicting results are found in the increasing the numbers of terminals (Krapp et al., literature, potentially due to the low species sampling 2014; Heller et al., 2015; Barfuss et al., 2016; Pinangé of Vriesea, which remains poorly representative of the et al., 2016; Schütz et al., 2016; Goetze et al., 2017; whole genus. To achieve a robust classification, the Gomes-da-Silva & Souza-Chies, 2017; Leme et al., last taxonomic revision of Tillandsioideae proposed a 2017; Maciel et al., 2018; Matuszak-Renger et al., series of rearrangements dividing large polyphyletic 2018). Nevertheless, large and complex to delimit genera into small, monophyletic, well-circumscribed genera such as Tillandsia L. (741 species), Pitcairnia morphological groups (Barfuss et al., 2016). Based L’Heritier (408 species), Vriesea Lindl. (226 species), on the stigma morphology, Vriesea s.l. was split into Puya Molina (226 species) and Ruiz & Pav. five new genera (Goudaea W.Till & Barfuss, Jagrantia (218 species) remain vastly under-sampled (Gouda, Barfuss & W.Till, Lutheria Barfuss & W.Till, Zizkaea

2 W.Till & Barfuss, Stigmatodon Leme, G.K.Br. & morphological and geographical variation of the Barfuss) and Vriesea s.s. (the Brazilian lineage) was species and type localities. To make the sampling as recognized as monophyletic. Finally, Cipuropsis Ule complete as possible, we also included specimens from was resurrected to accommodate the mesomorphic the living collections of the Marie Selby Botanical northern Andean ‘Vriesea’ spp. (Barfuss et al., 2016). Garden (SEL, United States), the Botanical Garden of However, the phylogenetic analysis conducted by the University of Vienna (WU-HBV, Austria), the Rio Gomes-da-Silva & Souza-Chies (2017) based on a de Janeiro Botanical Garden (RBvb, Brazil) and the total evidence approach (morphological data and four Jardin des Serres d’Auteuil (P, France). Details of the plastid markers) did not support the rearrangement of studied species and accession numbers are given in Vriesea s.l. proposed by Barfuss et al. (2016). Instead, the Supporting Information (Table S1). Gomes-da-Silva & Souza-Chies (2017) recognized Ananas comosus (L.) Merr. was included as an only two lineages: (1) Vriesea s.s. encompassing all outgroup for the different analyses. For the phyC Brazilian species including Stigmatodon, a new genus nuclear gene, sequences from additional six species described by Barfuss et al. (2016); and (2) Vriesea clade (including Ananas) where obtained from GenBank: ß composed partly by the Cipuropsis–Mezobromelia Ananas comosus (NCBI number KU095956.1), L.B.Sm. complex and by Josemania, Goudaea, Lutheria Guzmania wittmackii (André) André ex Mez (NCBI and Jagrantia, new genera described in Barfuss et al. number KX753898.1), Lutheria splendens (Brongn.) (2016). Thus, a consensus on genus delimitation in Barfuss & W.Till (NCBI number KX753915.1), Vrieseeae, particularly in Vriesea s.l., has not yet been Tillandsia heubergeri Ehlers (NCBI number reached, calling for additional systematics studies of KX753920), Tillandsia stricta Sol. KX753950.1) this group. and Tillandsia tenuifolia L. (NCBI number Here, we use the opportunity offered by next- KX753909.1). For the genome skimming, raw reads generation sequencing to produce a plastid dataset from a whole genome sequencing of Ananas comosus obtained using the genome skimming approach. ‘N67-10’ (NCBI number DRX020985) were added to We combined this plastid dataset with sequences the analyses together with the reads obtained in from the nuclear gene phyC to reconstruct the this study. phylogenetic tree of core Tillandsioideae. We test the monophyly of the genera in Vrieseeae, investigate phylogenetic relationships at the intergeneric level DNA EXTRACTION and within Vriesea s.s. and discuss the efficiency of Total genomic DNA was isolated from silicagel-dried genome skimming for shallow-level phylogenetic leaves collected in the field or taken from living reconstruction in bromeliads. collections using the Quiagen DNAeasy Plant Mini Kit (Qiagen, USA). We adjusted the manufacturer’s protocol to optimize the DNA extraction from the thick MATERIAL AND METHODS and fibrous leaves of bromeliads (contact authors for more details). Total DNA samples were evaluated TAXON SAMPLING with agarose gels and a NanoDrop Spectrophotometer Our dataset included 206 specimens (148 species) (Thermo Scientific, Waltham, MA, USA). The DNA

http://doc.rero.ch for plastid data and 171 specimens (134 species) for content was quantified with a Qubit Fluorometer v.2.2 nuclear data, totalling 227 specimens representing (Thermo Fisher Scientific, Waltham, MA, USA). 151 species and three infraspecific taxa from 11 genera of Tillandsioideae. Of these, 110 species NUCLEAR PHYC GENE AMPLIFICATION, ASSEMBLY AND belonged to Vriesea s.l., representing 49% of the total number of described species according to Gouda et al. ALIGNMENT (cont. updated). Besides Vriesea, the number of species The nuclear phyC region amplification was based on sampled in this study and the number of species per primers and protocols described by Louzada et al. genus according to Gouda et al. (cont. updated) were (2014) and Barfuss et al. (2016). Sequencing reactions as follows: (E.Morren ex Mez) Harms were carried out with the same amplification primers (11/42), Goudaea (2/2), Guzmania (4/218), Lutheria using the sequencing service from Microsynth (2/4), Mezobromelia (1/5), Racinaea M.A.Spencer (Switzerland). Forward and reverse sequences were & L.B.Sm. (1/78), Stigmatodon (5/18), Tillandsia trimmed, edited and assembled using Geneious v.6.1.8 (8/741), J.R.Grant (3/93) and Zizkaea (1/1) (Kearse et al., 2012). Consensus sequences were (Supporting Information, Table S1). aligned using ClustalW implemented in Geneious Most samples were collected during fieldwork v.6.1.8 and individual gap positions were treated as in Brazil. We tried to cover the greatest possible missing data.

3 LIBRARY PREPARATION AND GENOME SKIMMING 2010). SNP calling was performed for the plastid SEQUENCING genome using UnifiedGenotyper of GATK v.3.6 For library preparation, we quantified the DNA and (McKenna et al., 2010) and the EMIT_ALL_SITES diluted the samples in a 100 µl solution containing option to recover both variant and invariant sites. 500 ng of DNA and fragmented with a Bioruptor SNP calling was not performed for the nuclear sonicator (Diagenode) to obtain fragments of 300– genome as the sequencing depth of the genome 700 bp. For samples with lower DNA content, we used skimming methods is not high enough to recover this 100 µl of available aliquot without any dilution. genomic region. Only sites with a quality Q > 20, < Library preparation was performed following de 50% missing data and with a minimum depth of 3× La Harpe et al. (2018). DNA clean-up, size selection, were retained using vcftools v.0.1.13 (Danecek et al., end-repair and the A-tailing steps were achieved with 2011). Fasta files were generated using vcf-tab- a KAPA LTP library preparation kit (Roche, Basel, to-fasta (https://github.com/JinfengChen/vcf-tab- Switzerland). Adaptor ligation and adaptor fill-in to-fasta). Sequences were aligned using ClustalW steps were based on Meyer & Kircher (2010). For the implemented in Geneious 6.1.8 (Kearse et al., 2012). majority of samples, an aliquot of 4 µl of the ligated The samples that had > 50% missing data were fragment solution was amplified for eight cycles removed from the alignment. using the KAPA HiFi DNA Polymerase (Roche, Basel, Switzerland) and the set of 60 dual-index primers SUBSAMPLING OF PLASTID GENES designed by (Loiseau et al., 2019). Dual-indexed primers were chosen to avoid inaccuracies in multiplex Most phylogenetic studies in Bromeliaceae are based sequencing (Kircher, Sawyer & Meyer, 2012). Samples on a few plastid genes and/or nuclear genes (Louzada with a low amount of DNA were amplified for 12 et al., 2014; Evans et al., 2015; Aguirre-Santoro, cycles using 11 µl of the ligated fragment solution. The Michelangeli & Stevenson, 2016; Barfuss et al., 2016; products of library amplification were quantified using Gomes-da-Silva & Souza-Chies, 2017; Kessous et al., a Qubit Fluorometer v.2.2 (Thermo Fisher Scientific, 2019), resulting in poor resolution in some genera. Waltham, MA, USA). Genomic DNA libraries were Increasing the molecular sampling effort would help pooled equimolarly and sequenced in an Illumina in elucidating some of the hypotheses or resolving HiSeq 3000 Genome Analyzer (Illumina, San Diego, discordances (Kessous et al., 2019). Next-generation California, USA) in 1.5 lanes using 2 × 150 bp paired- sequencing (NGS), and especially genome skimming end reads at the University of Bern. that allows the sequencing of the whole plastid genome, represent great opportunities to overcome molecular data limitations in phylogenetic studies (Coissac DNA QUALITY, ASSEMBLY AND ALIGNMENT et al., 2016). However, NGS methods have high cost Quality control, quality score per base, sequence and different challenges at the library preparation, duplication level and overrepresented sequences were sequencing and bioinformatics steps. For this reason, checked with FASTQC (https://www.bioinformatics. we tested if the use of a limited number of plastid babraham.ac.uk/projects/fastqc/). To remove genes would be sufficient to obtain high phylogenetic sequencing errors, the reads were trimmed with support within Vriesea. For this purpose, we performed http://doc.rero.ch CONDETRI v.2.2 (Smeds & Künstner, 2011) using 20 the annotation of the partial plastome and selected the as high-quality threshold parameter. genes with a length > 900 bp (similar to the length For each sample, all reads were mapped to the usually obtained using Sanger sequencing) from the Tillandsia adpressiflora Mez pseudo-reference genome genome skimming alignment. Annotation for the built in de La Harpe et al. (2018) using BOWTIE2 partial plastome alignment obtained for one species v.2.2.5 (Langmead & Salzberg, 2012) and the ‘--very- (Vriesea marceloi) was performed in Geneious v.6.1.8 sensitive-local’ option. This pseudo-reference genome (Kearse et al., 2012) using Tillandsia usneoides L. as a was build using the high-quality and annotated reference genome (NCBI number KY293680.1, Poczai Ananas comosus reference genome (Ming et al., 2015) & Hyvönen, 2017). Then, a set of 18 high-quality genes as a start point. The method consisted of incorporating with sequence length greater than 900 bp was selected specific variation of T. adpressiflora into the Ananas and extracted from the annotated sequence. This genome to improve the mapping efficiency of samples selection included genes normally used in phylogenetic of Tillandsioideae. This pseudo-reference contains the studies with Bromeliaceae and also described as plastid genome. variables in Poczai & Hyvönen (2017). To ensure the data quality, reads were realigned To ensure that each region extracted from the around indels, and the base quality of reads was annotated sequence of V. marceloi Versieux & re-calibrated using GATK v.3.6 (McKenna et al., T.Machado corresponded to the selected genes,

4 sequences were checked using nucleotide BLAST RESULTS (Altschul et al., 1990) available at NCBI (https://www. GENOME SKIMMING ncbi.nlm.nih.gov/). After a quality check, each selected gene sequence was used as a reference to extract the The genome skimming approach generated corresponding region from the plastid alignment of all 358 443 406 high-quality reads (Q > 20; including our samples. Then, the alignments extracted for each mitochondrial, nuclear and plastid reads) across the gene were checked again using nucleotide BLAST 206 samples successfully sequenced in this study. (Altschul et al., 1990), available at NCBI (https://www. After mapping, SNP calling and quality filtering, an ncbi.nlm.nih.gov/), to confirm the equivalence with the alignment of 77 836 high-quality plastid bases was selected genes. obtain for the 206 samples. Only 3.6% of missing data were observed in the alignment and 75 654 bp were obtained on average per sample. The final alignment (including gaps after adding outgroups) was 78 483 bp PHYLOGENETIC ANALYSES and contained 6212 variable characters of which 2849 We performed phylogenetic inference using three were potentially parsimony-informative. To avoid datasets: (1) a concatenated unpartitioned alignment ambiguity about our plastid genome data set, we will of the plastome alignment generated by the genome refer to it as ‘partial plastome’ from here on. skimming approach; (2) the nuclear gene phyC and (3) a concatenated alignment of 18 selected plastid genes SUBSAMPLING OF PLASTID GENES AND NUCLEAR PHYC and the nuclear phyC (18 plastid genes + phyC) to evaluate whether a reduced set of genes would have GENE success in recovering well-supported trees. The latter The partial plastome annotated of V. marceloi featured was partitioned by genes and the best nucleotide the four typical plastid regions: one large single copy substitution model for each gene was selected using (LSC 56 433 bp), inverted repeat A (IRa 3197 bp), one JModelTest v.2.1.7 (Darriba et al., 2012) based on small single copy (SSC 13 259 bp) and inverted repeat the Bayesian information criterion (BIC) (Table B (IRb 4947 bp). The IRa and IRb were different sizes 1). The numbers of variable sites, conserved sites since the percentage of coverage and the recovered and potentially parsimony-informative sites were genes in each of these parts were not the same. Ninety calculated in MEGA 7 (Kumar, Stecher & Tamura, genes were recovered for V. marceloi, 35 totally and 2016) for all alignments (Table 1). 55 partially. Eighteen of these were selected and Maximum-likelihood (ML) analyses were performed extracted from the partial plastome alignment and using RAxML v.8.2.10 (Stamatakis, 2014) with 1000 they are shown in Table 1. rapid bootstrap replicate searches (Stamatakis, Hoover The phyC alignment for the 171 samples was 1006 bp & Rougemont, 2008). The partial plastome alignment, long with 297 variable characters of which 159 were nuclear phyC and the concatenated 18 plastid genes potentially parsimony-informative. The concatenated + phyC were analysed using GTRGAMMA as a alignment of the 18 plastid genes + phyC had a total nucleotide substitution model. length of 31 241 bp, 2323 of which were variable and Additionally, we applied a Bayesian inference (BI) 1205 potentially parsimony-informative (Table 1). For approach to phyC and to the concatenated alignment the calculation of conserved and variable sites, gaps http://doc.rero.ch of 18 plastid genes + phyC, using MrBayes v.3.2.6 were not considered. All sequences generated in this (Ronquist et al., 2012). As the size of the dataset (> 200 work were included in a free and online repository species and > 70 000 bp) exceeds the computational (NCBI, National Center for Biotechnology Information) limits of MrBayes, it was not possible to conclude number SUB6280204 for the raw read data and see the Bayesian analysis for the partial plastome. For Supporting Information, Table S1 for phyC sequences. each gene, we applied the best model of substitution selected by JModelTest (Table 1). Two independent PHYLOGENETIC RELATIONSHIPS runs of four Markov chains and 25 million generations were performed with sampling every The phylogenetic tree based on the partial plastome 1000 generations. All phylogenetic analyses were dataset (Fig. 2) had higher bootstrap support (BS) performed on the CIPRES Science Gateway (Miller, and recovered more clades than the one obtained Pfeiffer & Schwartz, 2011). Convergence among the from the phyC alone (Supporting Information, Fig. two runs was assessed in Tracer v.1.6.0 (Rambaut & S1) and from 18 plastid genes + phyC (Supporting Drummond, 2003). The consensus phylogenetic tree Information, Fig. S2). The main clades (tribes and node posterior probabilities were computed from Tillandsieae and Vrieseeae, subtribes Cipuropsidinae the posterior distribution of trees after removal of a and Vrieseinae) were recovered with high support on burn-in of 25%. trees obtained from partial plastome and 18 plastid

5 Table 1. Matrix allignment statistics. Recovery percentage is calculated in comparison with the genome used as reference for plastid annotation T. usneoides (Poczai & Hyvönen, 2017). For the calculation of conserved and variable sites, gaps were not considered.

Length of Recovery Number of Nunber of Number of potentially Substitution alignment % conserved variable parsimony-informative model sites (%) sites (%) sites (%)

atpA 1527 100 1461 (96) 66 (4) 32 (2) HKY+I atpB 1086 75 1047 (96) 39 (4) 16 (1) HKY+I atpF 1395 100 1286 (92) 109 (8) 54 (4) F81+I+G clpP 951 46 868 (91) 80 (8) 37 (4) HKY+I ndhA 2060 96 1886 (92) 168 (8) 94 (5) HKY+I ndhD 1230 81 1142 (93) 73 (6) 32 (3) HKY+G psaA 1945 87 1889 (97) 56 (3) 23 (1) HKY+I psaB 1016 46 954 (94) 54 (5) 41 (4) HKY+I+G psbA 1008 95 980 (97) 28 (3) 15 (1) HKY+I psbB 1246 81 1190 (96) 40 (3) 26 (2) HKY+I rpl16 1221 82 1139 (93) 77 (6) 44 (4) HKY+I+G rpoB 2153 67 2038 (95) 105 (5) 58 (3) GTR+I+G rpoC1 2204 79 2053 (93) 134 (6) 75 (3) HKY+I+G rpoC2 2172 55 2007 (92) 165 (8) 74 (3) HKY+I trnK-UUU 1838 68 1633 (89) 203 (11) 100 (5) GTR+I+G trnl-GAU 906 89 895 (99) 11 (1) 2 (0.2) K80 ycf1 4813 85 4171 (87) 563 (12) 297 (6) HKY+I+G ycf3 1464 73 1386 (95) 67 (5) 30 (2) HKY+I phyC 1006 _ 693 (69) 297 (30) 159 (15) SYM+I+G All genes 31241 _ 28730 (92) 2323 (7) 1205 (4) GTR+G concatenated Partial plastome 78483 49 71623 (91) 6212 (8) 2849 (4) GTR+G

genes + phyC. However, the tree obtained from phyC Vrieseinae, containing Alcantarea, Vriesea s.s. and recovered Cipuropsidinae as sister to Tillandsieae Stigmatodon, were also recovered as monophyletic with (BS 65/PP 0.95) and both as a sister group of strong support (BS 99, Fig. 2). Although Alcantarea was Vrieseinae (BS 92/PP 0.89). The phyC tree recovered monophyletic with strong support (BS 100), Vriesea extremely low support values within Vriesea (BS < 50 s.l. is polyphyletic. A group of species was nested in and PP < 0.50). For the tree obtained with 18 plastid Stigmatodon (BS 100) and one species (V. lutheriana) genes + phyC, some clades were found in Vriesea, clustered with Tillandsia. The separation between but in general, the groups received weak support Vriesea and Stigmatodon was strongly supported with http://doc.rero.ch (Supporting Information, Fig. S2). Therefore, only a BS value of 100 (Fig. 2). Vriesea drepanocarpa (Baker) the results from the partial plastome are presented Mez (Fig. 1G) was recovered as sister to all other species and discussed here (Fig. 2); the phylogenetic trees of Vriesea s.s.. Within this group, 12 main clades were of the phyC and 18 plastid genes + phyC genes are recovered with moderate-strong support in the partial shown in the Supporting Information (Figs S1, S2). plastome tree (BS values ranging from 80 to 100) (Fig. Tillandsieae were recovered with strong support (BS 2). Many Vriesea spp., including V. longicaulis (Baker) 100, Fig. 2) in a sister position to Vrieseeae, also strongly Mez, V. itatiaiae Wawra, V. medusa Versieux and supported (BS 100, Fig. 2). Guzmania is monophyletic V. ensiformis (Vell.) Beer, that were represented by more (BS 99) and sister to a clade composed of Tillandsia, than two accessions were not recovered as monophyletic. Racinaea and Vriesea lutheriana J.R.Grant (BS 100, Fig. 2). Cipuropsidinae were strongly supported (BS 100, Fig. 2) and contained two clades also supported by DISCUSSION a BS value of 100. The first was formed by Werauhia GENOME SKIMMING EFFICIENCY AND PHYLOGENETIC (BS 99) and Lutheria (BS 100). The second was composed of Zizkaea, the non-monophyletic Goudaea UTILITY and the lineages of the Cipuropsis–Mezobromelia The genome skimming approach uses Illumina complex (BS 100). technology to obtain high-copy fractions of the genome

6 http://doc.rero.ch

Figure 1. Species of the Vriesea s.s. clade showing habitat and inflorescence. A, . B, Vriesea medusa. C, Vriesea minor. D, Vriesea diamantinensis. E, Vriesea clausseniana. F, Vriesea gradata. G, Vriesea drepanocarpa. H, Vriesea pseudoatra. I, . J, . K, Vriesea hydrophora. L, Vriesea unilateralis. M, Vriesea longistaminea. N, Vriesea brusquensis. Photographs: T.M. Machado.

7 http://doc.rero.ch

8 http://doc.rero.ch

9 in low coverage, making it possible to retrieve sequence also found when analysing intrageneric relationships data for almost whole plastid genomes (Straub et al., in Pitcairnioideae s.l. using phyC, especially in relation 2012). When combined with a multiplex protocol (Meyer to Puya and Dyckia Schult.f. (Jabaily & Sytsma, 2010; & Kircher, 2010; Kircher et al., 2012), it allows multiple Krapp et al., 2014; Schütz et al., 2016). samples to be sequenced in a single run and thus to The concatenated matrix of 18 plastid genes + phyC obtain a large amount of genomic data at an affordable resulted in trees with moderate support. Some of cost (Straub et al., 2012; Malé et al., 2014; Dodsworth, the major clades recovered by partial plastome were 2015). Using this approach, we were able to recover also recovered by this concatenated dataset, but the nearly half of the plastid genome for 206 species of relationship between clades remained unclear due to Tillandsioideae, to our knowledge representing an lack of statistical support. Many of the selected genes unprecedented genomic dataset for an evolutionary were not fully recovered by genome skimming, but study in Bromeliaceae. The fact that we did not recover some showed a considerable number of potentially > 49% (an average of 75 654 bp) of the plastid genome parsimony-informative and variable sites. compared to the reference genome of the most closely Comparing the results of the trees obtained with related species, T. usneoides (159 657 bp; Poczai & partial plastome data and the concatenated 18 plastid Hyvönen, 2017) may be explained by the stringent genes + phyC matrix, it is evident that the genome quality filtering to retain a position in our study and skimming approach is advantageous due to the large by some extended deletions observed in our samples number of data obtained, in terms of both the number compared to the reference genome. Some authors of base pairs and the number of species that can be suggest that a high level of multiplexing may reduce multiplexed. In the case of our data, for which only the sequencing capacity by up to 10–30% (Straub et al., partial plastomes were recovered, we were able to 2012). In our case, we obtain a high sequencing coverage reconstruct phylogenetic trees that clarify intergeneric for the plastid (36× on average per sample) and only relationships with strong statistical support. However, 3.5% of missing data in our alignment of 77 836 bp. for intrageneric relationships, the relationships The continuous advances in Illumina sequencing between the recovered clades still lack statistical technologies now providing hundreds of millions of support. Other studies have shown that genome reads per lane and the option of using more than one skimming may be effective in clarifying intrageneric Illumina lane of sequencing allowed us to obtain high- relationships if few species are multiplexed, allowing quality data while multiplexing > 200 samples. a greater genome coverage (Parks, Cronn & Liston, Bromeliaceae show a low substitution rate (Smith & 2009; Kane et al., 2012; Malé et al., 2014; Dodsworth, Donoghue, 2008) and, in general, phylogenetic analyses 2015). In taxa that diverged recently, as is the case produce trees with low resolution (Sass & Specht, 2010; for most species of Bromeliaceae (Givnish et al., 2011, Maia et al., 2012; Versieux et al., 2012; Evans et al., 2015; 2014), whole genome sequencing or target sequencing Palma-Silva et al., 2016). Thus, phylogenetic studies that produces hundreds of nuclear markers would be of Bromeliaceae currently have sought more robust the next step to clarify intrageneric relationships. resolution by increasing the number of plastid markers and including nuclear markers (Krapp et al., 2014; PHYLOGENETIC RELATIONSHIPS IN CORE Aguirre-Santoro et al., 2016; Barfuss et al., 2016; Palma- TILLANDSIOIDEAE Silva et al., 2016). The phyC is the most used nuclear http://doc.rero.ch gene in phylogenetic studies of Bromeliaceae (Silvestro Petal appendages were a traditional morphological et al., 2014; Louzada et al., 2014; Barfuss et al., 2016; character used by Smith & Downs (1977) to differentiate Castello et al., 2016; Schütz et al., 2016; Goetze et al., Tillandsia from Vriesea. However, ontogenetic 2017). In our dataset, 15% of all characters of phyC were sequences revealed that petal appendages are the last potentially parsimony-informative, being the second external multicellular structures formed during the most variable among the selected genes (Table 1). development (Brown & Terry, 1992). Several studies However, when analysed separately, it generated trees have demonstrated the fragility of this character to with strong support only for intergeneric relations, differentiate genera due to high levels of homoplasy whereas within Vriesea support values were extremely (Schulte & Zizka, 2008; Barfuss et al., 2016; Gomes-da- low. A large polytomy and clades with low support were Silva & Souza-Chies, 2017). Nevertheless, the transfer

Figure 2. part 1. Cladogram from the ML analysis based on the partial plastome alignment (78 483 bp) for 206 specimens of Tillandsioideae. Names of tribes, subtribes and genera follow the classification of Tillandsioideae proposed by Barfuss et al. (2016). Numbers above the branches represent ML and BS values. Only values > 50% are shown. See Supporting Information for trees based on the nuclear gene phyC and the concatenation of 18 plastid genes + phyC. In detail: phylogram with proportional branch lengths from the ML tree.

10 of Vriesea spp. with petal appendages to Tillandsia results do not support the circumscription of Barfuss proposed by Grant (1993, 1995, 2004) was corroborated et al. (2016) of Goudaea as a monophyletic genus. by the phylogenetic analyses of Barfuss et al. (2016) and In Vrieseinae, which comprise the eastern Brazilian Gomes-da-Silva & Souza-Chies (2017). In our analysis, lineages and were the focus of our taxonomic sampling, T. heliconioides Kunth, T. malzinei (E.Morren) Baker Alcantarea emerges as the sister group of Stigmatodon, (formerly included in Vriesea) and T. juncea (Ruiz & a new genus segregated from Vriesea (Barfuss et al. Pav.) Poir. represent Tillandsia subgenus Tillandsia. 2016) and Vriesea s.s. Intrageneric phylogenetic This group is sister to a well-supported clade composed relationships in Alcantarea were generally well- of one species representative of the Tillandsia supported (Fig. 2). Although our sampling of this genus gardneri Lindl. complex, two species of Tillandsia is limited, evolutionary relationships seem to be linked subgens Anoplophytum (Beer) Baker and Vriesea to the geographical distribution since sister species lutheriana. The phylogenetic proximity of V. lutheriana tend to occur in geographical proximity [e.g. A. burle- with Tillandsieae highlights that, despite recent marxii (Leme) J.R.Grant and A. pataxoana Versieux; taxonomical rearrangements, Vriesea as currently A. glaziouana (Leme) J.R.Grant and A. martinellii circumscribed is still polyphyletic. Vriesea lutheriana Versieux & Wand.], a pattern already highlighted by was described by Grant (1992: 199) as a species Versieux et al. (2012) for the species from the Serra from with tripinnate inflorescences and dos Órgãos mountain range. Related species showing conduplicate-spiral stigma type. These morphological grouped geographical distribution pattern may be the characteristics are closely related to that of V. duidae result of forest fragmentation leading to geographical (L.B.Sm.) Gouda and V. rubrobracteata Rauh that have isolation of ancestral populations during the cycles of never been sequenced and V. elata (Baker) L.B.Sm. climatic changes of the Plio-Pleistocene and has been and V. zamorensis (L.B.Sm.) L.B.Sm. that were both hypothesized as a putative driver of speciation in recovered in the Cipuropsis–Mezobromelia complex this genus (Benzing, 2000; Versieux et al., 2012). Our sensu Barfuss et al. (2016). In addition to V. lutheriana, result indicates several Vriesea spp. [V. freicanecana the tripinnate inflorescence type is also known from J.A.Siqueira & Leme, V. lancifolia (Baker) L.B.Sm., six Tillandsia spp. from (León & Alva, 2008). Two V. oligantha (Baker) Mez and V. vellozicola Leme & of these, T. hildae Rauh and T. ferreyrae L.B.Sm., were J.A.Siqueira] cluster within the recently described sampled in Barfuss et al. (2016) and formed a well- genus Stigmatodon (Barfuss et al., 2016). These supported clade with T. heliconioides and T. malzinei species, which are epiphytes on species of Vellozia in Tillandsia subgenus Tillandsia. Likewise, the Vand. or rupicolous plants on inselbergs and quartzite position of V. lutheriana with Tillandsia spp. revealed areas of the Espinhaço mountain range differ from the in our analyses is statistically supported, although other members of the Stigmatodon clade in several we cannot establish the exact phylogenetic position morphological characters (e.g. stigma, rosette and due to our low sampling of Tillandsieae; we suggest floral/inflorescence structure) and habitat preferences. further investigations of the relationships among the The taxonomic revision of Stigmatodon was performed tripinnate species (V. lutheriana, T. hildae, T. ferreyrae). by Couto (2017) who named the species of Vriesea In Cipuropsidinae, two well-supported main clades nested in this genus the Vriesea limae L.B.Sm. group. were identified. The first is composed of Werauhia Taxonomic rearrangements will be proposed for the

http://doc.rero.ch and Lutheria. The second clade includes the species re-circumscription of Stigmatodon as a monophyletic of Goudaea and Zizkaea, both newly described by genus (D. Couto, personal communication). Barfuss et al. (2016), and a clade called the Cipuropsis– Vriesea s.s. is considered to be the ‘true’ Vriesea Mezobromelia complex by Barfuss et al. (2016) that because V. psittacina (Hook.) Lindl., the type species is composed of the North-Andean species V. dubia of the genus, is found in this lineage (Barfuss et al., (L.B.Sm.) L.B.Sm., V. zamorensis, V. elata and 2016; Gomes-da-Silva & Souza-Chies, 2017). According Mezobromelia capituligera (Griseb.) J.R.Grant. This to Gomes-da-Silva & Souza-Chies (2017), Vriesea group deserves further attention in order to clarify s.s. contains species essentially distributed in the whether the type species of Cipuropsis (C. subandina Paraná dominion (mostly Atlantic Forest) with a few Ule) will emerge among the mesomorphic northern occurrences in the Chacoan dominion. However, the Andean ‘Vriesea’ spp. and whether they are related to increased taxonomic sampling of our analysis revealed Mezobromelia (Barfuss et al., 2016; Gomes-da-Silva & that extra-Brazilian species such as V. laxa (Griseb.) Souza-Chies, 2017). Finally, the two Goudaea spp. do Mez and V. speckmeieri W.Till from , not form a monophyletic group, as G. chrysostachys V. macrostachya (Bello) Mez from Puerto Rico and (E.Morren) W.Till & Barfuss was found to be more closely V. oxapampae Rauh from Peru, are also present in this related to the Cipuropsis–Mezobromelia complex than clade, suggesting that calling this group of species the to G. ospinae (H.Luther) W.Till &Barfuss. Therefore, our ‘Brazilian lineage’ is potentially misleading. Despite

11 their geographical disjunction, these four species in misapplied names in herbarium specimens (Costa share a stigma convolute-blade type (Grant, 1997; et al., 2014). Some of these problematic species Till, 2008), a character referred to by Gomes-da-Silva have been recently revised (Costa, Rodrigues & & Souza-Chies (2017) as an important synapomorphy Wanderley, 2009; Kessous, Salgueiro & Costa, 2018; for the lineage. Thus, future research efforts in Vriesea Neves et al., 2018). For instance, the V. paraibica should focus on the sampling of Vriesea spp. from Wawra complex was treated taxonomically by Costa the Amazon, the Guiana Shield and the Andes to et al. (2009) and is represented in our study by clade investigate whether they belong to Vriesea s.s. or to I-B comprising V. duvaliana E.Morren, V. carinata the Cipuropsis–Mezobromelia complex. Wawra and V. gradata (Baker) Mez. However, the In summary, in our study Vriesea spp. emerged keels formed by two sepals, a synapomorphy for the in at least four evolutionary distinct lineages: (1) group (Costa et al., 2015), are missing in the last Tillandsia; (2) Cipuropsisinae; (3) Stigmatodon and species. Taxa in the V. incurvata Gaudich. complex, (4) Vriesea s.s. Based on this result, further taxonomic recently re-circumscribed (Neves et al., 2018), present rearrangements will be necessary to render Vriesea simple inflorescences and suberect peduncles bearing monophyletic. bracts similar to the floral ones. In our phylogenetic tree, this group was recovered with high support in clade I-C, but it did not form a clade since V. sucrei PHYLOGENETIC RELATIONSHIPS IN VRIESEA L.B.Sm. & Read was nested instead in clade I-A. The In Vriesea s.s., V. drepanocarpa (Baker) Mez was found latter clade combines the largest number of species in a sister position to the remaining species of the including the V. ensiformis complex treated by Kessous clade. This result contrasts with previous phylogenetic et al. (2018). Although clade I is composed mostly of studies based on morphological and molecular data, species with simple inflorescences, it also includes in which this species was nested in Tillandsia, in a four species with composed inflorescences and tubular sister position to Racinaea (Gomes-da-Silva & Souza- flowers [V. vagans (L.B.Sm.) L.B.Sm., V. brusquensis Chies, 2017). Unlike the other species of Vriesea, Reitz, V. philippocoburgii Wawra and V. schwackeana which have six to 12 columns of ovules in the locule, Mez] and one species with simple inflorescences and V. drepanocarpa only has four columns, similar to campanulate corolla type (V. wawranea Antoine). some species of Tillandsia and Racinaea (Kuhn et al., Finally, with the exception of V. schwackeana, which 2016). Therefore, the close position of V. drepanocarpa occurs in the Cerrado domain, all species in this clade to Tillandsia and Racinea in the study of Gomes- are mostly epiphytes (rarely terrestrial) distributed in da-Silva & Souza-Chies (2017) might be explained the Atlantic Forest. by the fact that they included this morphological Clade II comprises the taxa from the V. corcovadensis character in their phylogenetic analyses. In our ML Mez complex revised by Gomes-da-Silva & Costa (2011) phylogeny, a long branch separates V. drepanocarpa and Gomes-da-Silva et al. (2012), including V. poenulata from the rest of the Vriesea s.s. clade. This suggests (Baker) Mez, V. flammea L.B.Sm., V. lubbersii (Baker) that the morphological differentiation of this species E.Morren and V. triangularis Reitz. The monophyly of is underpinned by significant genetic divergence. the group was already questioned in a cladistic analysis Given that V. drepanocarpa also has a differentiated (Gomes-da-Silva et al., 2012), but further morphological and molecular analyses (Costa et al., 2015; Gomes-da-Silva

http://doc.rero.ch floral morphology with a simple-erect stigma type resembling that of Goudaea chrysostachys (Gomes- & Souza-Chies, 2017) confirmed that the V. corcovadensis da-Silva & Souza-Chies, 2017), further studies are complex is monophyletic. Our analysis confirms the needed to confirm the exact phylogenetic position of previous result since V. arachnoidea A.F.Costa was found this species. nested in a weakly supported clade related to clades VII Our study expands the sampling of Vriesea s.s. and VIII. Although our sample of V. arachnoidea was to species from other Brazilian biomes such as labelled as a topotype in the living collection where we the Caatinga and the Cerrado, as well as to extra- sampled it, we cannot exclude a misidentification to Brazilian species. Based on our plastid phylogenetic explain this unexpected placement in the phylogenetic tree, we define 12 main clades that have a good analysis. In addition, V. paratiensis E.Pereira clustered in statistical support within Vriesea s.s., and discuss clade II even though it does not have stolons, utriculiform their morphological or geographical characteristics. rosettes or linear-triangular blades that are considered as Clade I includes 31 species (Fig. 2) combining synapomorphies for this group. nearly all sampled species with flat and simple Clade III includes eight species divided into two inflorescences and tubular and distichous flowers, groups. The first one is composed of Vriesea penduliflora except for V. rhodostachys L.B.Sm. that was recovered L.B.Sm. and V. thyrsoidea Mez, two epiphytic, as sister to clade II. Most of these species have terrestrial or rupicolous species that grow as at complex morphological delimitations, often resulting elevations > 1500 m. Vriesea penduliflora is restricted

12 to the higher areas of the Mantiqueira mountain in herbarium collections as V. diamantinensis Leme range and is included in the Red List of the Brazilian and V. simulans Leme (clade VIII) due to their similar Flora (CNCFlora, 2012a), whereas V. thyrsoidea is morphologies. Furthermore, they also exhibit a great restricted to the higher parts of the Serra dos Orgãos. intraspecific morphological variation that renders Both species have compound inflorescences, primary their taxonomic delimitation challenging (pers. obs.) bracts covering the pedicel of the branches and (Fig. 1B, D). tubular flowers secund at anthesis (Smith & Downs, Clade VI includes species with simple inflorescences, 1977). The second group includes V. longicaulis, distichous flowers and campanulate corolla type. V. itatiaiae, V. stricta L.B.Sm., V. bituminosa Wawra Although they are morphologically closely related to and V. jonghei (K.Koch) E.Morren. Several samples clade V, there is no support in our phylogenetic trees for were included for each of these species and our a sister relationship among these two clades. Vriesea result indicates that none of them are monophyletic. macrostachya (Bello) Mez, a species distributed in the Vriesea sceptrum Mez emerged as sister of all other Greater Antilles (Moura, 2011) was recovered as sister species in clade III and is a species restricted to the to all other species of the clade. Despite their disjunct higher parts of the Mantiqueira mountain range. It distributions, V. macrostachya morphologically is morphologically similar to V. penduliflora with resembles V. chapadensis Leme, a species restricted to compound inflorescences, primary bracts covering the the rock outcrops in the northern part of the Espinhaço pedicel of the branches but with tubular flowers never mountain range (Moura, 2011). One species in this secund at anthesis (Machado & Menini Neto, 2010). clade (V. longicaulis) has a distinct morphology with Clade IV includes V. unilateralis (Baker) Mez, distichous floral bracts and flowers secund at anthesis. V. botafogensis Mez, V. oxapampae Rauh and However, the three samples of this species are placed V. saundersii (Carrière) E.Morren ex Mez. Vriesea in distinct clades, making the position of this species saundersii and V. botafogensis were also recovered as uncertain. member of a similar clade (i.e. clade R) in Gomes-da- Clades VII, VIII and IX are moderately supported Costa & Souza-Chies (2017). Vriesea botafogensis was clades containing 12 morphologically related species. considered as a synonym of V. saundersii until a decade The phylogenetic relationships among these three ago because these species have similar inflorescences, clades are weakly supported (Fig. 2). Most species flowers and rosettes and both are endemic to inselbergs present compound inflorescences with campanulate in the city of Rio de Janeiro (Smith & Downs, 1977; corolla type and flowers secund at anthesis. Although Leme & Costa, 1994). they share similarities regarding the inflorescence Clade V includes 17 species (Fig. 2) with simple or morphology, these clades are distributed in different compound inflorescences and campanulate corolla habitats. Although the species of clade VII are type, with the exception of Vriesea stricta L.B.Sm. epiphytes from the Atlantic Forest, species in Clade and V. jonesiana Leme that have tubular flowers. The VIII a rupicolous species of the Cerrado domain and floral morphology of most species in this clade was species of Clade IX are found at the transition between formerly used as a criteria to delimitate V. section this biome and the Atlantic Forest (BFG, 2018). Xiphion (E.Morren) E.Morren ex Mez, a group that Clade X includes six species: Vriesea neoglutinosa has been shown to be non-natural (Versieux et al., Mez, V. friburgensis Mez, V. procera (Mart. ex Schult. 2012; Costa et al., 2015; Gomes-da-Silva & Souza- & Schult.f.) Wittm., V. rodigasiana E.Morren and http://doc.rero.ch Chies, 2017). Two smaller groups stand out in clade V V. densiflora Mez. Vriesea densiflora has a tubular for their geographical structure. Clade V-A comprises rosette and congest inflorescences with tubular yellow V. pulchra Leme & L.Kollmann, V. linharesiae Leme flowers. This species is rupicolous and is an endemic & J.A.Siqueira, V. carmeniae R.Moura & A.F.Costa of rock outcrops in the Diamantina Plateau region and V. minutiflora Leme, all restricted to either the along the central portion of the Espinhaço mountain north-eastern region of Brazil or the Caatinga domain range in the Cerrado domain (Versieux & Wendt, 2006; at elevations > 800 m, except for V. pulchra that also Versieux et al., 2010). The placement of V. densiflora occurs in the Atlantic Forest. Clade V-B includes the in clade X, outside the ‘V. minarum L.B.Sm. group’, sympatric species V. medusa (Fig. 1B) and V. nanuzae proposed by Versieux (2011), is surprising given its Leme, restricted to montane areas in the Diamantina morphological and ecological divergence with the other Plateau region along the central portion of the species of this clade. All species except V. densiflora Espinhaço mountain range (Versieux, 2008; Versieux have infundibuliform rosettes, lax and compound et al., 2010). These two species show great morphological inflorescences and usually reddish, yellow and tubular similarity, both with compound and highly branched flowers with exserted stamens (except in V. procera inflorescences as well as campanulate flowers secund in which they are included) (Smith & Downs, 1977). at anthesis (Versieux, 2008; Leme, Trindade-Lima & They are epiphytic, rupicolous or terrestrial species Ribeiro, 2010). These species are often misidentified distributed along the Atlantic Forest in several

13 habitats of this domain as inselberg mats and restingas secund yellow flowers at anthesis with campanulate (BFG, 2018). Vriesea procera has a wide geographical corollas and exserted stamens. This floral morphology distribution with species reaching northern South suggests a bat pollination, whereas V. minarum and America, and is divided into four varieties (Smith V. marceloi have tubular rosettes and yellow tubular & Downs, 1977). Our data show that V. procera, flowers secund at anthesis that are pollinated by V. neoglutinosa and V. friburgensis are related, and hummingbird, suggesting that these floral characters they have been recognized as the V. procera complex are labile in this group (Versieux, 2011; Versieux due to the difficulty in recognizing and delimiting taxa & Machado, 2012). These species were considered in this group (Versieux & Wendt, 2006; Costa, Gomes- morphologically related to V. stricta (clades III and V) da-Silva & Wanderley, 2014; Uribbe, 2014). and V. densiflora (clade X) in taxonomic works, but our Clade XI comprises V. sincorana Mez and genomic data placed them in different clades. V. atropurpurea Silveira, which share similarities Many Vriesea spp. show considerable intraspecific in the rosette and inflorescences and have flowers morphological variation, and this is often insufficiently with campanulate corollas and exserted stamens documented in morphological descriptions and poorly (Moura, 2011). Both species are restricted to the represented in identification keys (Costa et al., 2009; Espinhaço mountain range, but V. atropurpurea Neves et al., 2018). To address these complex species occurs to the south in the Cerrado domain, whereas delimitations, we included more than one sample per V. sincoranea occurs to the north in the Caatinga species in our phylogenetic analysis. Our results show domain (Moura, 2011; BFG, 2018). There is a striking that many of these species are not monophyletic. For phenotypic similarity between V. atropurpurea instance, two samples (547 and 550) of V. medusa collected and V. longstaminea C.C.Paula & Leme that leads in the same locality were found in distant positions in the to frequent misidentifications in herbaria (Leme phylogenetic tree. Likewise, the samples of V. itatiaiae & Paula, 2004; Coffani-Nunes et al., 2010; Moura, (704 and 574) from the same locality did not appear as a 2011). They were recovered as sister species in the monophyletic lineage. These findings could be explained study of Gomes-da-Silva & Souza-Chies (2017) and either by the poor resolution between clades in Vriesea were treated as synonyms in Moura (2011). Vriesea s.s. due to the lack of informative characters at a low level longistaminea (Fig. 1M) is restricted to an area of < 8 of divergence. Alternatively, the non-monophyly of many km2 in an ironstone outcrop region known as the Iron species of Vriesea could reflect true biological processes. Quadrangle (Leme & Paula, 2004; Jacobi et al., 2007) Indeed, species complexes and non-monophyletic species and is categorized as Critically Endangered (CR) in are common in plants (Naciri & Linder, 2015; Pinheiro, the Red List of the Brazilian Flora (CNCFlora, 2012b). Dantas-Queiroz & Palma-Silva, 2018) and are thought to Vriesea atropurpurea and V. longistaminea do not occur be the result of hybridization or incomplete lineage sorting. sympatrically but their distributions are separated Assuming that post-zygotic barriers for reproduction of by only 100 km. Despite their strong morphological bromeliads are potentially weak (Palma-Silva et al., 2011; similarity and geographic proximity, our analysis Wagner et al., 2015) and that the individuals often grow in suggests that these species are not phylogenetically mixed aggregates of species sharing the same pollinator, closely related since V. longistaminea is sister to clade interspecific gene flow may occur and lead to the formation VIII. Therefore, the morphological similarity between of natural hybrids (Palma-Silva et al., 2011; Lexer et al., 2016; Zanella et al., 2016; Neri, Wendt & Palma-Silva, http://doc.rero.ch V. atropurpurea and V. longistaminea is probably a case of convergent adaptation to similar environmental 2017; Mota et al., 2019). Furthermore, natural hybrids of conditions. In a large genus such as Vriesea, occupying closely related Vriesea spp. have been identified (Zanella contrasting environments including rock outcrops et al., 2016; Neri et al., 2017). Therefore, we hypothesize and ombrophilous forests, identifying occurrences of that hybridization could be one of the evolutionary phenotypic convergence can help in elucidating the process driving phenotypic variation and blurring species processes involved in the diversification of the group. delimitation in the genus. Clade XII includes L.B.Sm., V. clausseniana (Baker) Mez and V. marceloi Versieux & T.M.Machado, all rupicolous and heliophytic species CONCLUSIONS growing at > 1000 m elevation along the southern portion of the Espinhaço mountain range (Versieux Our study provided the first phylogenetic hypothesis et al., 2008). Specimens of V. marceloi can be found for Vrieseinae based on a comprehensive sampling in herbaria identified as V. clausseniana and the two of Vriesea and an extensive coverage of the plastid species occur sympatrically, although V. marceloi only genome. The genome skimming approach used in this grows above 1900 m of elevation where mist formation study allowed to recover large-scale plastid data to infer is frequent (Versieux & Machado, 2012). Vriesea the evolutionary relationships in core Tillandsioideae clausseniana has infundibuliform rosettes, and strongly with good support. Our results are congruent with

14 the taxonomic rearrangements proposed by Barfuss Almeida VR, da Costa AF, Mantovani A, Gonçalves- et al. (2016), with the exception of Goudaea, which was Esteves V, de Oliveira Arruda R do C, Forzza RC. 2009. not recovered as monophyletic. We show that Vriesea Morphological phylogenetics of Quesnelia (Bromeliaceae, remains polyphyletic, as suggested in previous works, a Bromelioideae). Systematic Botany 34: 660–672. finding that calls for further taxonomic rearrangements Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ. based in phylogenetic relationships and in-depth 1990. Basic local alignment search tool. Journal of Molecular morphological studies. More specifically, this would Biology 215: 403–410. include: the transference of V. lutheriana to Tillandsia; Barfuss MHJ, Samuel R, Till W, Stuessy TF. 2005. the transfer of several taxa to Stigmatodon; and a revision Phylogenetic relationships in subfamily Tillandsioideae of Cipuropsis. Our analysis does not corroborate the (Bromeliaceae) based on DNA sequence data from seven plastid regions. American Journal of Botany 92: 337–351. delimitation of Vriesea s.s. proposed by Gomes-da-Silva Barfuss MHJ, Till W, Leme EM, Pinzón JP, & Souza-Chies (2017), since we recovered Stigmatodon Manzanares JM, Halbritter H, Samuel R, Brown GK. as a separate lineage, whereas it was nested in Vriesea 2016. Taxonomic revision of Bromeliaceae subfam. s.s. in their study. Furthermore, we showed that Vriesea Tillandsioideae based on a multi-locus DNA sequence s.s. is not restricted to eastern Brazil; it also contains phylogeny and morphology. Phytotaxa 279: 1–97. species distributed in the Andes, the Caribbean and in Benzing DH. 2000. Bromeliaceae: profile of an adaptive other southern South American countries. In Vriesea radiation. New York: Cambridge University Press. s.s., 12 strongly supported clades are recognized and BFG- The Brazilian Flora Group. 2015. Growing knowledge: supported by morphological characters or geography. an overview of seed plant diversity in Brazil. Rodriguésia 66: Therefore, our phylogenetic study of Vriesea contributes 1085–1113. to overcome some of the limitations of the traditional BFG- The Brazilian Flora Group. 2018. Brazilian Flora based solely on morphological characters 2020: innovation and collaboration to meet Target 1 of the where well-defined morphological groups of species may Global Strategy for Plant Conservation (GSPC). Rodriguésia include convergent, yet unrelated, taxa. Finally, our 69: 1513–1527. work provides the basis for the selection of Vriesea s.s. Brown GK, Terry RG. 1992. Petal appendages in species to be used in future microevolutionary studies, Bromeliaceae. American Journal of Botany 79: which will aim at a better understanding of the drivers 1051–1071. of the evolution of this important floristic component of Castello LV, Barfuss MHJ, Till W, Galetto L, Chiapella JO. the Brazilian Atlantic Forest. 2016. Disentangling the Tillandsia capillaris complex: phylogenetic relationships and taxon boundaries in Andean populations. Botanical Journal of the Linnean Society 181: ACKNOWLEDGEMENTS 391–414. CNCFlora. 2012a. Vriesea penduliflora. In: Lista Vermelha The authors thank the Botanical Gardens of the University da flora brasileira versão 2012.2. Rio de Janeiro: Centro of Vienna (WU-HBV), Botanical Garden of Rio de Janeiro Nacional de Conservação da Flora. Available at: http:// (RBvb), Marie Selby Botanical Garden (SEL) and Jardin cncflora.jbrj.gov.br/portal/pt-br/profile/Vriesea penduliflora. des Serres d’Auteuil (P) for providing plant material and Accessed 1 April 2019. ICMBIO and IEF-MG for collection permits. TMM thanks CNCFlora. 2012b. Vriesea longistaminea. In: Lista Vermelha all the amazing people who collaborated in the fieldwork da Flora Brasileira versão 2012.2. Rio de Janeiro: http://doc.rero.ch and other stages of this work. TMM was supported Centro Nacional de Conservação da Flora. Available at: by PhD fellowship grants of CAPES (Coordenação de http://cncflora.jbrj.gov.br/portal/ pt-br/profile/Vriesea Aperfeiçoamento de Pessoal de Nível Superior) and longistaminea. Accessed 10 March 2019. CNPq (Conselho Nacional de Desenvolvimento Científico Coffani-Nunes JV, Versieux LM, Wanderley M das GL, e Tecnológico – CNPq-SWE (205660/2014-2)/CNPq Pirani JR. 2010. Flora da Serra do Cipó, Minas Gerais, (142354/2016-3)). LMV thanks CNPq (455510-2014-8 and Brasil: Bromeliaceae-Tillandsioideae. Boletim de Botânica 304778-2013-3). NS and CL received funding from the 28: 35–54. Swiss National Science Foundation (CRSII3-147630) and Coissac E, Hollingsworth PM, Lavergne S, Taberlet P. NS from the University of Lausanne. 2016. From barcodes to genomes: extending the concept of DNA barcoding. Molecular Ecology 25: 1423–1428. Costa AF, Gomes-da-Silva J, Wanderley M das GL. 2014. Vriesea (Bromeliaceae, Tillandsioideae): taxonomic history, REFERENCES and morphology of the Brazilian lineage. The Journal of the Aguirre-Santoro J, Michelangeli FA, Stevenson DW. Torrey Botanical Society 141: 338–353. 2016. Molecular phylogenetics of the Ronnbergia alliance Costa AF da, Gomes-da-Silva J, Wanderley M das GL. (Bromeliaceae, Bromelioideae) and insights into their 2015. Vriesea (Bromeliaceae, Tillandsioideae): a cladistic morphological evolution. Molecular Phylogenetics and analysis of eastern Brazilian species based on morphological Evolution 100: 1–20. characters. Rodriguésia 66: 429–440.

15 Costa AF da, Rodrigues PJFP, Wanderley M das GL. with description of two new species. Systematic Botany 36: 2009. Morphometric analysis and taxonomic revision of 291–309. the Vriesea paraibica complex (Bromeliaceae). Botanical Gomes-da-Silva J, Souza-Chies TT. 2017. What actually Journal of the Linnean Society 159: 163–181. is Vriesea? A total evidence approach in a polyphyletic Couto D. 2017. Revisão taxonômica e filogenia de Stigmatodon genus of Tillandsioideae (Bromeliaceae, ). Cladistics Leme, G.K. Br. & Barfuss (Bromeliaceae - Tillandsioideae). 34:181–199. Unpublished D. Phil. Thesis, Universidade Federal do Rio de Gomes-da-Silva J, Vargens FAC, Arruda RCO, Costa AF. Janeiro. 2012. A morphological cladistic analysis of the Vriesea Crayn DM, Winter K, Smith JAC. 2004. Multiple origins corcovadensis group (Bromeliaceae: Tillandsiodeae), with of Crassulacean acid metabolism and the epiphytic habit anatomical descriptions: new evidence of the non-monophyly in the Neotropical family Bromeliaceae. Proceedings of the of the genus. Systematic Botany 37: 641–654. National Academy of Sciences 101: 3703–3708. Gouda EJ, Butcher D, Gouda CS. cont. updated. Danecek P, Auton A, Abecasis G, Albers CA, Banks E, Encyclopaedia of bromeliads, version 4. Utrecht: University De Pristo MA, Handsaker RE, Lunter G, Marth GT, Botanic Gardens. http://bromeliad.nl/encyclopedia/. Accessed Sherry ST. 2011. The variant call format and VCFtools. 20 February 2019. Bioinformatics 27: 2156–2158. Grant JR. 1992. Vriesea lutheriana, a new species with Darriba D, Taboada GL, Doallo R, Posada D. 2012. tripinnate inflorescence from Costa Rica. Journal of the jModelTest 2: more models, new heuristics and parallel Bromeliad Society (USA) 42:114–117. computing. Nature Methods 9: 772. Grant JR. 1993. True misplaced in Vriesea Dodsworth S. 2015. Genome skimming for next-generation (Bromeliaceae: Tillandsioideae). Phytologia 75: 170–175. biodiversity analysis. Trends in Plant Science 20: 525–527. Grant JR. 1995. New combinations and new taxa in the Escobedo-Sarti J, Ramírez I, Leopardi C, Carnevali G, Bromeliaceae. Phytologia 79: 254–256. Magallón S, Duno R, Mondragon D. 2013. A phylogeny Grant JR. 1997. The identity of the West Indian endemic of Bromeliaceae (Poales, Monocotyledoneae) derived from an Vriesea macrostachya (Bromeliaceae). Nordic Journal of evaluation of nine supertree methods. Journal of Systematics Botany 17: 157–159. and Evolution 51: 743–757. Grant JR. 2004. New combinations and names in Andean Evans TM, Jabaily RS, de Faria APG, de Sousa L de OF, Pitcairnia, Tillandsia, and Werauhia. Vidalia 2: 23–25. Wendt T, Brown GK. 2015. Phylogenetic relationships in Heller S, Leme E, Schulte K, Benko-Iseppon AM, Zizka G. Bromeliaceae subfamily Bromelioideae based on chloroplast 2015. Elucidating phylogenetic relationships in the DNA sequence data. Systematic Botany 40: 116–128. alliance: AFLP analysis of Portea and the Gravisia complex Gilmartin AJ. 1973. Transandean distributions of (Bromeliaceae, Bromelioideae). Systematic Botany 40: Bromeliaceae in . Ecology 54: 1389–1393. 716–725. Givnish TJ, Barfuss MHJ, Ee BV, Riina R, Schulte K, Horres R, Zizka G, Kahl G, Weising K. 2000. Molecular Horres R, Gonsiska PA, Jabaily RS, Crayn DM, phylogenetics of Bromeliaceae: evidence from trnL(UAA) Smith JAC, Winter K, Brown GK, Evans TM, Holst BK, intron sequences of the chloroplast genome. Plant Biology 2: Luther H, Till W, Zizka G, Berry PE, Sytsma KJ. 2011. 306–315. Phylogeny, adaptive radiation, and historical biogeography Jabaily RS, Sytsma KJ. 2010. Phylogenetics of Puya in Bromeliaceae: insights from an eight-locus plastid (Bromeliaceae): placement, major lineages, and evolution of phylogeny. American Journal of Botany 98: 872–895. Chilean species. American Journal of Botany 97: 337–356. Givnish TJ, Barfuss MHJ, Ee BV, Riina R, Schulte K, Jacobi CM, Do Carmo FF, Vincent RC, Stehmann JR.

http://doc.rero.ch Horres R, Gonsiska PA, Jabaily RS, Crayn DM, 2007. Plant communities on ironstone outcrops: a diverse Smith JAC, Winter K, Brown GK, Evans TM, Holst BK, and endangered Brazilian ecosystem. Biodiversity and Luther H, Till W, Zizka G, Berry PE, Sytsma KJ. 2014. Conservation 16: 2185–2200. Adaptive radiation, correlated and contingent evolution, Kane N, Sveinsson S, Dempewolf H, Yang JY, Zhang D, and net species diversification in Bromeliaceae. Molecular Engels JMM, Cronk Q. 2012. Ultra-barcoding in cacao Phylogenetics and Evolution 71: 55–78. (Theobroma spp.; Malvaceae) using whole chloroplast Givnish TJ, Millam KC, Berry PE, Sytsma KJ. genomes and nuclear ribosomal DNA. American Journal of 2007. Phylogeny, adaptive radiation, and historical Botany 99: 320–329. biogeography of Bromeliaceae inferred from ndhF Kearse M, Moir R, Wilson A, Stones-Havas S, Cheung M, sequence data. Aliso: A Journal of Systematic and Sturrock S, Buxton S, Cooper A, Markowitz S, Duran C. Evolutionary Botany 23: 3–26. 2012. Geneious Basic: an integrated and extendable desktop Goetze M, Zanella CM, Palma-Silva C, Büttow MV, software platform for the organization and analysis of Bered F. 2017. Incomplete lineage sorting and hybridization sequence data. Bioinformatics 28: 1647–1649. in the evolutionary history of closely related, endemic Kessous IM, Neves B, Couto DR, Paixão-Souza B, yellow-flowered Aechmea species of subgenus Ortgiesia Pederneiras LC, Moura RL, Barfuss MH, Salgueiro F, (Bromeliaceae). American Journal of Botany 104: 1073–1087. Costa AF. 2019. Historical biogeography of a Brazilian Gomes-da-Silva J, Costa AF da. 2011. A taxonomic revision of lineage of Tillandsioideae (subtribe Vrieseinae, group (Bromeliaceae: Tillandsioideae) Bromeliaceae): the Paranaean Sea hypothesized as the main

16 vicariant event. Botanical Journal of the Linnean Society Loiseau O, Olivares I, Paris M, de La Harpe M, Weigand A, 192: 625–641. Koubinova D, Rolland J, Bacon CD, Balslev H, Kessous IM, Salgueiro F, Costa AF. 2018. Nomenclature and Borchsenius F, Cano A, Couvreur TLP, Delnatte C, epitypification of Vriesea (Bromeliaceae) species described in Fardin F, Gayot M, Mejia F, Machado TM, Perret M, Vellozo’s Flora fluminensis. Phytotaxa 350: 24–32. Roncal J, Sanin MJ, Stauffer F, Lexer C, Kessler M, Kircher M, Sawyer S, Meyer M. 2012. Double indexing Salamin N. 2019. Targeted capture of hundreds of nuclear overcomes inaccuracies in multiplex sequencing on the genes unravels phylogenetic relationships of the diverse Illumina platform. Nucleic Acids Research 40: e3. Neotropical palm tribe Geonomateae. Frontiers in Plant Krapp F, Pinangé DS de B, Benko-Iseppon AM, Science 10: 864. Leme EMC, Weising K. 2014. Phylogeny and evolution of Louzada RB, Schulte K, Wanderley M das GL, Silvestro D, Dyckia (Bromeliaceae) inferred from chloroplast and nuclear Zizka G, Barfuss MHJ, Palma-Silva C. 2014. Molecular sequences. Plant Systematics and Evolution 300: 1591–1614. phylogeny of the Brazilian endemic genus Orthophytum Krömer T, Kessler M, Herzog SK. 2006. Distribution and (Bromelioideae, Bromeliaceae) and its implications on flowering ecology of bromeliads along two climatically morphological character evolution. Molecular Phylogenetics contrasting elevational transects in the Bolivian Andes. and Evolution 77: 54–64. Biotropica 38: 183–195. Machado TM, Forzza RC, Stehmann JR. 2016. Bromeliaceae Kuhn SA, Nogueira FM, Fagundes NF, Mariath JE. 2016. from Caparaó National Park, Minas Gerais/Espírito Santo Morphoanatomy of the ovary and ovule in Bromeliaceae states, Brazil, with notes on distribution and conservation. subfamily Tillandsioideae and its systematic relevance. Oecologia Australis 20:133–146. Botanical Journal of the Linnean Society 181: 343–361. Machado TM, Menini Neto L. 2010. Bromeliaceae de um Kumar S, Stecher G, Tamura K. 2016. MEGA7: molecular campo de altitude no sul de Minas Gerais (Brasil). Fontqueria evolutionary genetics analysis version 7.0 for bigger datasets. 56: 109–124. Molecular Biology and Evolution 33: 1870–1874. Maciel JR, Louzada RB, Benko-Iseppon AM, Zizka G, de La Harpe M, Hess J, Loiseau O, Salamin N, Lexer C, Alves M. 2018. Polyphyly and morphological convergence Paris M. 2018. A dedicated target capture approach in Atlantic Forest species of Aechmea subgenus Chevaliera reveals variable genetic markers across micro- and macro- (Bromeliaceae). Botanical Journal of the Linnean Society evolutionary time scales in palms. Molecular Ecology 188: 281–295. Resources 19: 221–234. Maia VH, Da Mata CS, Franco LO, Cardoso MA, de La Harpe M, Paris M, Hess J, Barfuss MH, Serrano- Cardoso SRS, Hemerly AS, Ferreira PCG. 2012. DNA Serrano ML, Ghatak A, Chaturvedi P, Weckwerth W, barcoding Bromeliaceae: achievements and pitfalls. PLoS Till W, Salamin N. 2018. Genomic footprints of repeated One 7: e29877. evolution of CAM photosynthesis in tillandsioid bromeliads. Malé PJG, Bardon L, Besnard G, Coissac E, Delsuc F, bioRxiv 495812. Engel J, Lhuillier E, Scotti-Saintagne C, Tinaut A, Langmead B, Salzberg SL. 2012. Fast gapped-read alignment Chave J. 2014. Genome skimming by shotgun sequencing with Bowtie 2. Nature Methods 9: 357–359. helps resolve the phylogeny of a pantropical tree family. Leme EM, Heller S, Zizka G, Halbritter H. 2017. New Molecular Ecology Resources 14: 966–975. circumscription of Cryptanthus and new cryptanthoid Martinelli G, Valente ASM, Maurenza D, Kutschenko DC, genera and subgenera (Bromeliaceae: Bromelioideae) based Judice DM, Silva DS, Fernandez EP, Martins EM, on neglected morphological traits and molecular phylogeny. Barros FSM, Sfair JC. 2013. Avaliações de risco de Phytotaxa 318: 1–88. extinção de espécies da flora brasileira. In: Martinelli G,

http://doc.rero.ch Leme EM, Paula CC. 2004. Two new species of Brazilian Moraes MA, eds. Livro vermelho da Flora do Brasil. Rio de Bromeliaceae. Vidalia 2: 21–29. Janeiro: Andrea Jacobsson Estúdio, Instituto de Pesquisas Leme EM, Trindade-Lima T, Ribeiro OB de C. 2010. Jardim Botânico, 60–84. Revision of the lithophytic Vriesea species from Minas Gerais Martinelli G, Vieira CM, Gonzalez M, Leitman P, State, Brazil–Part IV. Journal of the Bromeliad Society 60: Piratininga A, da Costa AF, Forzza RC. 2008. 17–31. Bromeliaceae da Mata Atlântica brasileira: lista de espécies, Leme EMC, Costa A. 1994. e V. botafogensis distribuição e conservação. Rodriguésia 59: 209–258. duas espécies distintas. Bromélia 1: 11–18. Matuszak-Renger S, Paule J, Heller S, Leme EM, León B, Alva AS. 2008. General overview of Tillandsia Steinbeisser GM, Barfuss MH, Zizka G. 2018. subgenus Tillandsia in Peru: the three-pinnate species and Phylogenetic relationships among Ananas and related taxa the case of two endemic species. Revista Peruana de Biología (Bromelioideae, Bromeliaceae) based on nuclear, plastid and 15: 25–30. AFLP data. Plant Systematics and Eevolution 304: 841–851. Lexer C, Marthaler F, Humbert S, Barbará T, McKenna A, Hanna M, Banks E, Sivachenko A, de la Harpe M, Bossolini E, Paris M, Martinelli G, Cibulskis K, Kernytsky A, Garimella K, Altshuler D, Versieux LM. 2016. Gene flow and diversification in Gabriel S, Daly M. 2010. The genome analysis toolkit: a a species complex of Alcantarea inselberg bromeliads. MapReduce framework for analyzing next-generation DNA Botanical Journal of the Linnean Society 181: 505–520. sequencing data. Genome Research 20: 1297–1303.

17 Meyer M, Kircher M. 2010. Illumina sequencing library and evolution: a review of South American studies. Critical preparation for highly multiplexed target capture and Reviews in Plant Sciences 37: 54–80. sequencing. Cold Spring Harbor Protocols 2010: prot5448- Poczai P, Hyvönen J. 2017. The complete chloroplast genome pdb.prot5448. sequence of the CAM epiphyte Spanish moss (Tillandsia Miller MA, Pfeiffer W, Schwartz T. 2011. The CIPRES usneoides, Bromeliaceae) and its comparative analysis. PLoS science gateway: a community resource for phylogenetic One 12: e0187199. analyses. In: Proceedings of the 2011 TeraGrid Conference: Rambaut A, Drummond AJ. 2003. Tracer: MCMC trace Extreme Digital Discovery. Salt Lake City: ACM. analysis tool. Available at: http://tree.bio.ed.ac.uk/software/ Ming R, VanBuren R, Wai CM, Tang H, Schatz MC, tracer/5. Bowers JE, Lyons E, Wang ML, Chen J, Biggers E, Ronquist F, Teslenko M, Van Der Mark P, Ayres DL, Zhang J, Huang L, Zhang L, Miao W, Zhang J, Ye Z, Darling A, Höhna S, Larget B, Liu L, Suchard MA, Miao C, Lin Z, Wang H, Zhou H, Yim WC, Priest HD, Huelsenbeck JP. 2012. MrBayes 3.2: efficient Bayesian Zheng C, Woodhouse M, Edger PP, Guyot R, Guo HB, phylogenetic inference and model choice across a large model Guo H, Zheng G, Singh R, Sharma A, Min X, Zheng Y, space. Systematic Biology 61: 539–542. Lee H, Gurtowski J, Sedlazeck J, Harkess A, Sass C, Specht CD. 2010. Phylogenetic estimation of the McKain MR, Liao Z, Fang J, Liu J, Zhang X, Zhang Q, core bromelioids with an emphasis on the genus Aechmea Hu W, Qin Y, Wang K, Chen LY, Shirley N, Lin YR, Liu LY, (Bromeliaceae). Molecular Phylogenetics and Evolution 55: Hernandez AG, Wright CL, Bulone V, Tuskan GA, 559–571. Heath K, Zee F, Moore PH, Sunkar R, Leebens-Mack JH, Schulte K, Zizka G. 2008. Multi locus plastid phylogeny of Mockler T, Bennetzen JL, Freeling M, Sankoff D, Bromelioideae (Bromeliaceae) and the taxonomic utility of petal Paterson AH, Zhu X, Yang X, Smith JAC, Cushman JC, appendages and pollen characters. Candollea 63: 209–225. Paull RE, Yu Q. 2015. The pineapple genome and the Schütz N, Krapp F, Wagner N, Weising K. 2016. evolution of CAM photosynthesis. Nature Genetics 47: 1435. Phylogenetics of Pitcairnioideae s.s. (Bromeliaceae): evidence Mota MR, Pinheiro F, Leal BSS, Wendt T, Palma-Silva C. 2019. from nuclear and plastid DNA sequence data. Botanical The role of hybridization and introgression in maintaining species Journal of the Linnean Society 181: 323–342. integrity and cohesion in naturally isolated inselberg bromeliad Silvestro D, Zizka G, Schulte K. 2014. Disentangling populations. Plant Biology 21: 122–132. the effects of key innovations on the diversification of Moura RL. 2011. Revisão taxonômica do grupo Vriesea Bromelioideae (Bromeliaceae). Evolution 68: 163–175. platynema Gaudich. (Bromeliaceae). Unpublished D. Phil. Smeds L, Künstner A. 2011. ConDeTri-a content dependent Thesis, Universidade Federal do Rio de Janeiro. read trimmer for Illumina data. PLoS One 6: e26314. Naciri Y, Linder HP. 2015. Species delimitation and Smith LB, Downs RJ. 1974. Pitcairnioideae (Bromeliaceae). relationships: the dance of the seven veils. Taxon 64: 3–16. Flora Neotropica Monographs 14: 1–662. Neri J, Wendt T, Palma-Silva C. 2017. Natural hybridization Smith LB, Downs RJ. 1977. Tillandsioideae (Bromeliaceae). and genetic and morphological variation between two Flora Neotropica Monographs 14: 663–1492. epiphytic bromeliads. AoB Plants 10: plx061. Smith SA, Donoghue MJ. 2008. Rates of molecular evolution Neves B, Uribbe FP, Jacques SSA, Zanella CM, are linked to life history in flowering plants. Science 322: da Costa AF. 2018. Species boundaries in the Vriesea 86–89. incurvata (Bromeliaceae) complex after a broad morphometric Stamatakis A. 2014. RAxML version 8: a tool for and taxonomic study. Systematic Botany 43: 870–888. phylogenetic analysis and post-analysis of large phylogenies. Palma-Silva C, Leal BSS, Chaves CJN, Fay MF. 2016. Bioinformatics 30: 1312–1313.

http://doc.rero.ch Advances in and perspectives on evolution in Bromeliaceae. Stamatakis A, Hoover P, Rougemont J. 2008. A rapid Botanical Journal of the Linnean Society 181: 305–322. bootstrap algorithm for the RAxML web servers. Systematic Palma-Silva C, Wendt T, Pinheiro F, Barbará T, Fay MF, Biology 57: 758–771. Cozzolino S, Lexer C. 2011. Sympatric bromeliad species Straub SC, Parks M, Weitemier K, Fishbein M, Cronn RC, (Pitcairnia spp.) facilitate tests of mechanisms involved in Liston A. 2012. Navigating the tip of the genomic iceberg: species cohesion and reproductive isolation in Neotropical next-generation sequencing for plant systematics. American inselbergs. Molecular Ecology 20: 3185–3201. Journal of Botany 99: 349–364. Parks M, Cronn R, Liston A. 2009. Increasing phylogenetic Terry RG, Brown GK, Olmstead RG. 1997a. Phylogenetic resolution at low taxonomic levels using massively parallel relationships in subfamily Tillandsioideae (Bromeliaceae) sequencing of chloroplast genomes. BMC Biology 7: 1. using ndhF sequences. Systematic Botany 22: 333–345. Pinangé DS, Krapp F, Zizka G, Silvestro D, Leme EM, Terry RG, Brown GK, Olmstead RG. 1997b. Examination Weising K, Benko-Iseppon AM. 2016. Molecular of subfamilial phylogeny in Bromeliaceae using comparative phylogenetics, historical biogeography and character sequencing of the plastid locus ndhF. American Journal of evolution in Dyckia (Bromeliaceae, Pitcairnioideae). Botany 84: 664–664. Botanical Journal of the Linnean Society 183: 39–56. Till W. 2008. Vriesea speckmaieri–a new lower-altitude cloud Pinheiro F, Dantas-Queiroz MV, Palma-Silva C. 2018. forest species from northern Venezuela. Journal of the Plant species complexes as models to understand speciation Bromeliad Society 58: 250–255.

18 Uribbe FP. 2014. Variação morfológica no complexo Vriesea Versieux LM, Machado TM. 2012. A new ornithophilous procera (Mart. ex Schult. & Schult. f. (Tillandsioideae, yellow-flowered Vriesea (Bromeliaceae) from Serra do Bromeliaceae). Unpublished Master’s Thesis, Universidade Caraça, Minas Gerais, Brazil. Phytotaxa 71: 36–41. Federal do Rio de Janeiro. Versieux LM, Wendt T. 2006. Checklist of Bromeliaceae of Versieux LM. 2008. Checklist and one new species of Minas Gerais, Brazil, with notes on taxonomy and endemism. Bromeliaceae from Pico do Itambé, Minas Gerais, Brazil. Selbyana 27: 107–146. Botanical Journal of the Linnean Society 158: 709–715. Versieux LM, Wendt T. 2007. Bromeliaceae diversity and Versieux LM. 2011. Brazilian plants urgently needing conservation in Minas Gerais state, Brazil. Biodiversity and conservation: the case of Vriesea minarum (Bromeliaceae). Conservation 16: 2989–3009. Phytotaxa 28: 35–49. Versieux LM, Wendt T, Louzada RB, Wanderley MGL. 2008. Versieux LM, Barbará T, Wanderley M das GL, Bromeliaceae da Cadeia do Espinhaço. Megadiversidade 4: Calvente A, Fay MF, Lexer C. 2012. Molecular 98–110. phylogenetics of the Brazilian giant bromeliads Wagner ND, Wöhrmann T, Öder V, Burmeister A, ( Alcantarea, Bromeliaceae): implications for Weising K. 2015. Reproduction biology and chloroplast morphological evolution and biogeography. Molecular inheritance in Bromeliaceae: a case study in Fosterella Phylogenetics and Evolution 64: 177–189. (Pitcairnioideae). Plant Systematics and Evolution 301: Versieux LM, Louzada RB, Viana PL, Mota N, 2231–2246. Wanderley MGL. 2010. An illustrated checklist of Zanella CM, Palma-Silva C, Goetze M, Bered F. 2016. Bromeliaceae from Parque Estadual do Rio Preto, Minas Hybridization between two sister species of Bromeliaceae: Gerais, Brazil, with notes on phytogeography and one new Vriesea carinata and V. incurvata. Botanical Journal of the species of Cryptanthus. Phytotaxa 10: 1–16. Linnean Society 181: 491–504.

SUPPORTING INFORMATION Additional Supporting Information may be found in the online version of this article at the publisher's web-site: Figure S1. Cladogram from the maximum-likelihood (ML) analysis based on nuclear gene phyC for 171 specimens of Tillandsioideae. First numbers above the branches represent bootstrap values (BS) and the second number represents posterior probabilities (PP). Only values > 50% are shown. In detail, phylogram with proportional branch lengths from the BI and ML analyses. Figure S2. Cladogram from the Bayesian inference (BI) based on 19 concatenated genes (atpA, atpB, atpF, clpP, ndhA, ndhD, psaA, psaB, psbA, psbB, rpl16, rpoB, rpoC1, rpoC2, trnK-UUU, trnl-GAU, ycf1, ycf3, phyC) for 206 species of Tillandsioideae. The first number above the branches represents PP, and the second number represents BS values. Only values > 50% are shown. In detail, phylogram with proportional branch lengths from the BI and ML analyses. Table S1. Studied material with silica number/NCBI sample names for plastid reads (SUB6280204), voucher, locality and GenBank accession numbers for nuclear gene phyC. Abbreviations: BHCB, Herbarium of Universidade Federal de Minas Gerais; R, Herbarium of Museu Nacional (Brazil); RBvb, Bromeliad Living Collection of the Botanical Garden of Rio de Janeiro; SEL, Marie Selby Botanical Garden; WU-HBV, Botanical Garden of the http://doc.rero.ch University of Vienna; P, Jardin des Serres d’Auteuil Botanical Garden.

19